-
Notifications
You must be signed in to change notification settings - Fork 6
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add Attention Test Suite #37
base: main
Are you sure you want to change the base?
Conversation
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
# query_type;key_type;value_type;scale_type | ||
type_and_layout_combinations=( | ||
"f16;f16;f16;f16" | ||
) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Want to test other combinations? What could we test?
Generally, the test suite can and should include tests that don't currently pass.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The current test suite has not supported yet the other combinations. It will support; however, it needs some discussions - that is one of my upcoming tasks.
# Generates a test function in the generated MLIR code. | ||
# The generated function will take the same arguments as iree_linalg_ext.attention variants | ||
# and will just call iree_linalg_ext.attention variants with them, returning its result. | ||
def generate_function( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Bringing back some discussion from iree-org/iree#17751 (comment), could this generator produce operations other than iree_linalg_ext.attention
? Fine to start with only that one, but what about iree_linalg_ext.online_attention
or even torch.nn.functional.scaled_dot_product_attention
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Better to ask this Kunwar @Groverkss.
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Signed-off-by: erman-gurses <[email protected]>
Progress on this #2
This PR refactors, adds, and migrates Attention e2e tests from the PR here: iree-org/iree#18320